Intel and Google TPU Rumor Sparks Debate as Analysts Point to Packaging, Not Full Chip Manufacturing
A recent report claiming that Intel may manufacture Google’s custom AI chips has drawn fresh attention to the fast-moving artificial intelligence hardware supply chain. The chips in question are Google’s tensor processing units, or TPUs, which are purpose-built processors designed to accelerate AI workloads.
The report suggested that Google had placed an order with Intel for as many as three million TPU chips. That claim quickly stirred discussion across the semiconductor industry, especially as demand for advanced AI chips continues to strain global manufacturing capacity.
However, analysts from JPMorgan and Citi have offered a more cautious interpretation. According to JPMorgan, the situation may not involve Intel fully manufacturing Google’s TPUs. Instead, the bank believes Intel’s role is more likely tied to advanced chip packaging, while the main chip fabrication would still be handled by TSMC.
The difference is important. In semiconductor production, fabrication refers to the process of manufacturing the actual silicon dies. Packaging, on the other hand, involves assembling multiple chip components together into a finished product. As AI processors become more complex, advanced packaging has become a critical part of performance, power efficiency, and supply chain flexibility.
Intel’s EMIB-T packaging technology appears to be at the center of the discussion. EMIB-T is viewed as a lower-cost alternative to TSMC’s high-end CoWoS packaging technology. CoWoS has become highly sought after because it is widely used for advanced AI accelerators, but demand has created capacity bottlenecks. That has encouraged major technology companies to explore alternative packaging suppliers.
JPMorgan pushed back against the idea that Intel had won full chip manufacturing work from Google, describing the report as offering little clearly new information. The bank noted that companies such as Google could be evaluating Intel as a backup option due to tight capacity at TSMC, but it said there is limited evidence that Intel is actually fabricating the TPU chips.
According to JPMorgan’s view, the relevant Google TPU components are still expected to be produced by TSMC. The compute die is believed to use TSMC’s 2-nanometer process, while the input-output die is expected to use the 3-nanometer process. Intel’s potential involvement would therefore be focused on packaging rather than producing the core silicon.
Citi offered a slightly broader interpretation. While the bank acknowledged that many investors believe the report is mainly about packaging, it also said the partnership could potentially involve Intel’s foundry services and design services in addition to EMIB-T packaging. That leaves room for Intel to play a larger role in future AI chip projects, even if the current situation is not a full manufacturing win.
The debate highlights how important advanced packaging has become in the AI chip race. As companies build more powerful custom processors, they often need to combine multiple dies, high-bandwidth memory, and specialized components in a single package. This makes packaging technologies such as Intel’s EMIB-T and TSMC’s CoWoS increasingly strategic.
For Intel, any deeper relationship with Google would be significant. The company has been working to expand its foundry business and position itself as a major alternative in the global semiconductor supply chain. Winning packaging work for a major AI customer would support that effort, even if it falls short of full chip fabrication.
For Google, diversifying its supply chain could help reduce dependence on a single manufacturing and packaging partner. As AI infrastructure demand grows, access to reliable production capacity is becoming just as important as chip design itself.
The key takeaway is that the Intel-Google TPU story may be more about advanced packaging than a major shift in chip manufacturing. While the original report raised the possibility of Intel producing millions of Google AI chips, analysts currently see TSMC remaining central to the fabrication process. Intel, meanwhile, may benefit from growing demand for alternative packaging capacity as the AI hardware market continues to expand.
